feat: update rule;

core/config.py

@@ -18,6 +18,7 @@ MAX_SINGLE_CHUNK_SIZE = 5000
 META_KEY = "META"
 META_KEY = "META"
 DEFAULT_RULESET_ID = "通用"
 DEFAULT_RULESET_ID = "通用"
 ALL_RULESET_IDS = ["通用", "借款", "担保", "财务口", "金盘", "金盘简化", "麓发测试"]
 ALL_RULESET_IDS = ["通用", "借款", "担保", "财务口", "金盘", "金盘简化", "麓发测试"]
+MAX_WORKERS = 10
 
 
 use_lufa = False
 use_lufa = False
 if use_lufa:
 if use_lufa:
@@ -52,7 +53,6 @@ elif system == "Darwin":
 # docker设置
 # docker设置
 if use_docker:
 if use_docker:
     root_path = "/app"
     root_path = "/app"
-MAX_WORKERS = 20
 LLM = {
 LLM = {
     "base_tool_llm": LLMConfig(),
     "base_tool_llm": LLMConfig(),
     "fastgpt_segment_review": LLMConfig(
     "fastgpt_segment_review": LLMConfig(

core/tools/segment_rule_router.py

@@ -8,7 +8,7 @@ from core.tool import tool, tool_func
 from core.tools.segment_llm import LLMTool
 from core.tools.segment_llm import LLMTool
 
 
 
 
-ROUTER_SYSTEM_PROMPT = '''
+ROUTER_SYSTEM_PROMPT = """
 你是合同分段规则路由智能体（SegmentRuleRouter）。
 你是合同分段规则路由智能体（SegmentRuleRouter）。
 
 
 你的任务是：基于“当前分段文本”，从候选审查规则中选出“应执行审查”的规则项。
 你的任务是：基于“当前分段文本”，从候选审查规则中选出“应执行审查”的规则项。
@@ -26,10 +26,10 @@ ROUTER_SYSTEM_PROMPT = '''
 - 严格输出 JSON。
 - 严格输出 JSON。
 - 每个命中规则需给出简短 reason，说明该分段为何与规则相关。
 - 每个命中规则需给出简短 reason，说明该分段为何与规则相关。
 - 若确实没有任何相关规则，返回 {"selected_items": []}。
 - 若确实没有任何相关规则，返回 {"selected_items": []}。
-'''
+"""
 
 
 
 
-ROUTER_USER_PROMPT = '''
+ROUTER_USER_PROMPT = """
 【当前分段文本】
 【当前分段文本】
 {segment_text}
 {segment_text}
 
 
@@ -44,10 +44,10 @@ ROUTER_USER_PROMPT = '''
 
 
 【任务】
 【任务】
 请从候选规则中选择当前分段应执行的审查项，并输出 selected_items。
 请从候选规则中选择当前分段应执行的审查项，并输出 selected_items。
-'''
+"""
 
 
 
 
-ROUTER_OUTPUT_SCHEMA = '''
+ROUTER_OUTPUT_SCHEMA = """
 ```json
 ```json
 {
 {
   "selected_items": [
   "selected_items": [
@@ -58,7 +58,7 @@ ROUTER_OUTPUT_SCHEMA = '''
   ]
   ]
 }
 }
 ```
 ```
-'''
+"""
 
 
 
 
 @tool("segment_rule_router", "分段规则路由")
 @tool("segment_rule_router", "分段规则路由")
@@ -102,11 +102,7 @@ class SegmentRuleRouterTool(LLMTool):
 
 
     def _build_candidate_rules(self, rules: List[Dict]) -> List[Dict]:
     def _build_candidate_rules(self, rules: List[Dict]) -> List[Dict]:
         return [
         return [
-            {
-                r.get("title", ""): r.get("rule", "")
-            }
-            for r in rules
-            if r.get("title")
+            {r.get("title", ""): r.get("rule", "")} for r in rules if r.get("title")
         ]
         ]
 
 
     def _route_rules(
     def _route_rules(
@@ -120,28 +116,41 @@ class SegmentRuleRouterTool(LLMTool):
             return []
             return []
 
 
         candidates = self._build_candidate_rules(rules)
         candidates = self._build_candidate_rules(rules)
-        user_content = ROUTER_USER_PROMPT.format(
-            segment_text=segment_text,
-            context_memories_json=json.dumps(context_memories or [], ensure_ascii=False),
-            party_role=party_role,
-            candidate_rules_json=json.dumps(candidates, ensure_ascii=False),
-        ) + ROUTER_OUTPUT_SCHEMA
+        user_content = (
+            ROUTER_USER_PROMPT.format(
+                segment_text=segment_text,
+                context_memories_json=json.dumps(
+                    context_memories or [], ensure_ascii=False
+                ),
+                party_role=party_role,
+                candidate_rules_json=json.dumps(candidates, ensure_ascii=False),
+            )
+            + ROUTER_OUTPUT_SCHEMA
+        )
 
 
         llm_selected: List[Dict] = []
         llm_selected: List[Dict] = []
         try:
         try:
-            resp = self.run_with_loop(self.chat_async(self.build_messages(user_content)))
+            resp = self.run_with_loop(
+                self.chat_async(self.build_messages(user_content))
+            )
             data = self.parse_first_json(resp)
             data = self.parse_first_json(resp)
             llm_selected = data.get("selected_items", []) or []
             llm_selected = data.get("selected_items", []) or []
         except Exception:
         except Exception:
             llm_selected = []
             llm_selected = []
 
 
-        selected_titles = {str(item.get("title", "")).strip() for item in llm_selected if item.get("title")}
+        selected_titles = {
+            str(item.get("title", "")).strip()
+            for item in llm_selected
+            if item.get("title")
+        }
         selected_reasons = {
         selected_reasons = {
             str(item.get("title", "")).strip(): str(item.get("reason", "")).strip()
             str(item.get("title", "")).strip(): str(item.get("reason", "")).strip()
             for item in llm_selected
             for item in llm_selected
             if item.get("title")
             if item.get("title")
         }
         }
-        trigger_titles = self._match_trigger_titles(segment_text=segment_text, rules=rules)
+        trigger_titles = self._match_trigger_titles(
+            segment_text=segment_text, rules=rules
+        )
         merged_titles = selected_titles | trigger_titles
         merged_titles = selected_titles | trigger_titles
 
 
         if not merged_titles:
         if not merged_titles:

data/benchmark/eval.py

@@ -121,7 +121,7 @@ def _parse_args() -> argparse.Namespace:
     parser.add_argument(
     parser.add_argument(
         "--datasets-dir",
         "--datasets-dir",
         type=Path,
         type=Path,
-        default=base / "results" / "jp-output-lufa-20260416-000112",
+        default=base / "results" / "jp-output-lufa-20260416-235546",
         help="Directory containing Word files with annotations.",
         help="Directory containing Word files with annotations.",
     )
     )
     parser.add_argument(
     parser.add_argument(

utils/openai_util.py

@@ -4,6 +4,7 @@ from openai import AsyncOpenAI
 from dataclasses import dataclass
 from dataclasses import dataclass
 from tenacity import retry, stop_after_attempt, stop_after_delay, wait_fixed
 from tenacity import retry, stop_after_attempt, stop_after_delay, wait_fixed
 import asyncio
 import asyncio
+from core.config import MAX_WORKERS
 
 
 
 
 @dataclass
 @dataclass
@@ -12,8 +13,9 @@ class LLMConfig:
     api_key: str
     api_key: str
     model: str
     model: str
 
 
+
 class OpenAITool:
 class OpenAITool:
-    def __init__(self, llm_config: LLMConfig, max_workers: int = 5):
+    def __init__(self, llm_config: LLMConfig, max_workers: int = MAX_WORKERS):
         self.max_workers = max_workers
         self.max_workers = max_workers
         self.llm_config = llm_config
         self.llm_config = llm_config
         self.client = AsyncOpenAI(
         self.client = AsyncOpenAI(
@@ -24,17 +26,11 @@ class OpenAITool:
     async def chat(self, msg, tools=None):
     async def chat(self, msg, tools=None):
         if tools is None:
         if tools is None:
             extra_body = None
             extra_body = None
-            if msg[0]['role'] == 'system':
-                extra_body = {
-                    'variables': {
-                        'system': msg[0]['content']
-                    }
-                }
+            if msg[0]["role"] == "system":
+                extra_body = {"variables": {"system": msg[0]["content"]}}
                 msg = msg[1:]
                 msg = msg[1:]
             response = await self.client.chat.completions.create(
             response = await self.client.chat.completions.create(
-                model=self.llm_config.model, 
-                messages=msg,
-                extra_body=extra_body
+                model=self.llm_config.model, messages=msg, extra_body=extra_body
             )
             )
             content = response.choices[0].message.content
             content = response.choices[0].message.content
             reasoning_content = response.choices[0].message.model_extra.get(
             reasoning_content = response.choices[0].message.model_extra.get(
@@ -57,4 +53,4 @@ class OpenAITool:
             async with sem:
             async with sem:
                 return await self.chat(m, tools)
                 return await self.chat(m, tools)
 
 
-        return await asyncio.gather(*[_wrapped(m) for m in msgs])
+        return await asyncio.gather(*[_wrapped(m) for m in msgs])
\ No newline at end of file